AI in the Workplace: What 17 Million ChatGPT Messages Reveal (OpenAI Study)

On August 11, 2026, OpenAI published a landmark study on how organizations actually use generative AI: 1,500 companies, more than 17 million messages analyzed. Verdict: large companies are no longer experimenting — they've built AI into their daily workflows. And the lesson applies just as much to small businesses in the Basque Country.

Growth that's no longer anecdotal

The working paper How Organizations Use AI: Evidence from ChatGPT, authored by researchers from OpenAI, Columbia Business School and the Wharton School, cross-references ChatGPT Enterprise accounts with employee roles, task types and the financial data of publicly traded companies, all through March 2026.

First takeaway: adoption is exploding. Between June 2025 and March 2026, the number of ChatGPT Enterprise customer companies multiplied roughly sevenfold, and total token volume generated also multiplied sevenfold. This isn't a curiosity effect: once a company adopts, it deepens its usage month after month. The six-month sample analyzed covers more than 1,500 organizations and over 17 million messages.

The practical takeaway for a small business: generative AI is no longer a topic to keep an eye on — it's a standard production tool in the most structured organizations. Every month of delay widens the efficiency gap on administrative and communication tasks.

Who adopts first? The most structured companies

Second takeaway: among publicly traded US companies, adoption is concentrated among the largest, most highly valued, and most R&D- and sales-intensive companies. Companies in the top revenue quartile have about 6.9 percentage points higher adoption probability than average; those in the top 5% show a 9.8-point gap. Even comparing companies within the same sector, the gap remains 7.2 to 11.3 points.

This profile isn't surprising: these companies have teams that can experiment, structured data and budgets. But the important fact lies elsewhere: adoption correlates with organizational capacity, not size on its own. A small business that formalizes its processes (quotes, follow-ups, reports, content) can replicate this behavior without hiring a data team.

Junior staff are the first drivers of adoption

Third takeaway, and a counterintuitive one: active use spans every function and every level of the hierarchy, but with very uneven intensity. Early-career employees send roughly eight to nine more messages per week than their senior colleagues. Marketing and communications teams use the tool more intensively than leadership does. Managers and directors make up about 24% of weekly active users, executives and founders 11%, juniors and interns 7% — but in terms of message volume, the hierarchy flips.

For a small business owner, the conclusion is immediate: your youngest team members are the best adoption accelerators. Training them early, giving them a usage framework and collecting their feedback costs less than any top-down training program. Conversely, an organization where only leadership uses AI misses out on most of the actual work volume.

What companies actually do with AI

Fourth takeaway: usage spans a broad spectrum of intellectual work — writing, technical work, communication and information synthesis. The task mix varies by sector and function, but these four families come up everywhere. They're exactly the tasks that fill a small business's day-to-day: writing a quote, answering an email, updating a website, summarizing a meeting, preparing a sales proposal.

The study also shows that companies differ sharply in the speed, scale and purpose of their adoption: some automate deeply, others stick to writing assistance. The authors conclude that organizations are still learning to weave AI into their workflows — nobody is “done,” not even the giants.

What this changes for a small business in the Basque Country

Take a shopkeeper in Anglet, a tradesperson in Bayonne or a firm in Biarritz. Their critical tasks are writing and communication: customer replies, quotes, invoices, content. That's precisely what the study observes at the core of large organizations too. The difference is that a small business doesn't need to roll out an enterprise tool: an AI assistant properly configured on its own documents is enough to capture 80% of the gains.

In practice, a small business in the Basque Country can aim, within the first three months, for: customer replies drafted in two minutes instead of twenty, quotes generated from a validated template, automatic reports after every appointment, and a social media presence kept up without sacrificing fieldwork. All of it with a human validating — AI stays an assistant, not a replacement.

An example: the typical week of an automated freelancer

Take a tradesperson in the BAB area (Bayonne-Anglet-Biarritz) who installs kitchens. Their week revolves around the same tasks: replying to quote requests received by email or phone, visiting clients, drafting quotes, following up with prospects, keeping up social media. With an AI assistant configured on their standard documents, their week changes:

• Monday morning: quote requests are sorted automatically, a personalized response is prepared for each prospect, and the tradesperson approves it in ten minutes instead of an hour.

• After each visit: the report is dictated over the phone and automatically structured into the client's file, ready to be shared.

• The quote: generated from the validated template, with the measurements and options discussed on site; an automatic follow-up goes out if the client hasn't replied within eight days.

• Social media: photos from job sites are turned into weekly posts, reviewed and then scheduled.

This scenario isn't theoretical: each of these building blocks exists today, and OpenAI's study shows that this exact type of task is where real usage is concentrated in organizations. The tradesperson doesn't just save time: they improve their conversion rate, because every prospect gets a fast, polished response, even when the job site is keeping the owner busy.

Comparison: what should you choose based on your situation?

From assistants to agents: the next step

OpenAI's study measures usage where a human stays in the loop: an employee asks, the assistant answers, the human validates. That's the state of the art in 2026 for most organizations. But the trajectory is already visible: the same task families — writing, communication, synthesis — are moving from “assistant” mode to “agent” mode, meaning automations that trigger, execute and check part of the work without intervention at every step.

For a small business, this distinction is strategic. The assistant saves the human time; the agent makes the repetitive task disappear. In practice: instead of drafting every customer reply with ChatGPT, an agent can sort incoming requests, prepare a reply based on the company's documents, submit it for approval, then send it. The owner only checks in — and only when the matter deserves it.

The study's data suggests this shift will happen first in highly repetitive, low-risk tasks — emails, quotes, reports, content updates. All of these are within reach for a small business with the right support, no infrastructure or technical team required. The advantage for the Basque Country here is having a fabric of agile small businesses that can test these automations faster than large groups.

What the study doesn't say

An honest reading of this working paper also means seeing its limits. The sample of publicly traded companies is American; European and French coverage is marginal. The analysis focuses on ChatGPT Enterprise, a product aimed at organizations with a certain level of maturity — small businesses appear only indirectly. Finally, the authors say it themselves: the results are likely to evolve, and adoption remains uneven, even among comparable companies.

Two precautions follow for a small business. First, don't mechanically copy large-company practices: their context (teams, data, budgets) isn't yours. Second, don't wait for a “small business” study before acting: the four task families identified — writing, technical, communication, synthesis — can be measured in any small structure, no academic study required, in a week of observation.

5-step action plan

1. Map out one week of work: list the writing, communication and synthesis tasks that come up every week.

1. Choose a pilot scope: one single process (customer replies or reports) rather than a blanket rollout.

1. Give the tool to your youngest team members and collect their real usage data for two weeks.

1. Measure the time saved on the pilot before scaling up: the study confirms that usage intensity rises over time.

1. Formalize the rules: what data can go into the tool, who validates the outputs, how to archive them.

FAQ

Organizations that adopt generative AI are no longer pioneers: increasingly, they're just companies, full stop. The August 11, 2026 study demonstrates this with massive data. For a small business in the Basque Country, the question is no longer whether AI will become a routine work tool, but when it will start — and who will train their teams first.

Want to take action? Mister Anderson helps small and medium-sized businesses in the Basque Country make AI a practical reality: process audits, custom automations, team training. Let's talk about your project.